How to Figure Out What People Want

In my piece last week about five new thinking styles for working with thinking machines, I wrote about why it is important, in the age of AI, to think in terms of sequences rather than essences: learning to cope with dynamic contexts rather than trying to locate fixed, discoverable truths. This week I’m covering a practical application of this thinking style. 

Was this newsletter forwarded to you? Sign up to get it in your inbox.

In both of these cases, the obvious next question is: “How?”

A dominant framework for thinking about this question is to consider customer needs. Under this worldview, people simply buy the products they need. Or to put it another way, they “hire” products to do specific jobs for them. 

If you use this framework, you often implicitly adopt the view that people have an objective, discoverable set of problems that they need solved. And that if you look closely enough, you can map out those problems, connect them to needs, and thereby identify products to build.

This is a scientific approach to the building of products. Something previously mysterious becomes an empirical activity: You go out and discover needs, and then create ways to fulfill them. This means talking to customers—on the street, through surveys, on the internet, in bars—however and wherever you can find them. 

A naive way to do this is to grab a potential customer by the collar and straight-up ask them, “What do you want?” Any experienced founder can tell you this doesn’t normally work. You get shrugs or, if you’re in Manhattan, the occasional, “A fuckin’ raise!” 

People don’t have a list of needs at hand. 

Instead, an experienced founder acts more like an ethnographer. Their goal is to build up a detailed understanding of their target customer: what they do when they wake up in the morning, how they go through their inbox, and how they respond when they see that their boss replied to their long, detailed, late-night email with the single word “No.” The idea is to be objective. You ask questions to gather neutral data. That data tells you the truth about the world. You don’t want to fool yourself with leading questions and trying to confirm your priors.

But as any experienced founder knows, this process is fractal.

The deeper you get into asking questions about a person’s day, the more detail there is to observe. And true needs or jobs—the ones that are burning or unfulfilled—get lost in the process of mapping the infinite contours of an individual life.

What actually works? Drop the objectivity. Instead, lead with a particular idea or bias. Once you present to a customer your own view of the world or a perspective on what you intend to build, everything changes. Suddenly, the squishy mess inside their head hardens into visibility. It’s like a non-Newtonian fluid subjected to force: a liquid until you press on it with the strength of your own perspective.

You’ve probably had an analogous experience in other areas of your life. Consider the question: “What do you want for dinner?” Or, “Which movie do you want to watch?” Or, “What kind of article do you want to read?”

For me, I draw a blank. 

But if you ask me about the food at Sukh or what I thought about Inside Out 2, I can and will unfurl an incredibly long string of sentences with minute detail about why Sukh is the best Thai in the city—the hor mok in particular—and how Inside Out 2 is catnip for therapy nerds with, by the way, a totally unsatisfying story arc.

Why the difference in my ability to respond? 

Needs arise in context—they are not context-free facts

When you search for needs or jobs in the way I outlined earlier, your underlying theory is that there is an objective set of hidden desires. You are looking for essences. You’re adopting the rationalist view that these inherent properties not only exist, but can be discovered and enumerated formally.

But what we find instead is that these needs or jobs arise out of the right sequence of actions. A need or a “job” comes about as a result of the right prompt from you. This prompt, or provocation, is a creative act. It is the process of learning to generate sequences of words and behaviors that create the kind of responses from prospective customers that you want. By trying to be objective, you actually lower the chances of observing what you hope to observe.

Let’s take as an example my writing this article. It would be silly of me to ask you what kind of article you wanted me to write ahead of time. Presumably, you read it because it appeared in your inbox, not because you had an underlying need for it that you would be able to articulate.

The need only arose because I formed a sequence that generated a response in you. This response only looks like a need you’ve had all along, but it didn’t emerge until the context was created for it to do so.

Now the question becomes: “How do you learn to generate the right sequences?”

Learn to be the right kind of person

If you’re thinking, “Figure out the kinds of sequences that generate good responses,” you’re still looking for essences. You’re seeking a list of words that can make someone excited.

Instead, the process of making something that people want is the process of learning, through experiment and error, to be the kind of person who can generate needs, wants, and jobs in other people. 

This kind of person is one who notices that a new restaurant in their neighborhood has a line out the door. Instead of walking by, they walk in. 

They stop to notice the soft, earthy color palette of its interior decoration, one that evokes a coastal Mediterranean village. They see the way its menu layers in unexpected ingredients like za’atar, cinnamon, and chile as subtle references to other cultures and traditions. They notice the feelings that this sequence of experience evokes in them, the way it feels familiar and also pleasantly surprising. They know that if they linger on these feelings, they’ll be able to evoke them later—for themselves and for others—in a logo design or an article.

A good metaphor for how to become this kind of person is to think about the way that language models learn to respond to a prompt. They try over and over to generate sequences, and they learn implicit patterns that they apply in particular contexts. By learning to pay attention, they build an intuition for how to respond in any given situation. 

Language models don’t have a dictionary of all of the possible sequences that they can produce. Instead, once they are trained, the appropriate responses just emerge from the prompt. You are similar—except you are constantly training, and new responses are constantly emerging.

Making something that people want is an alternate way of seeing the world, and of responding to it.

You learn it a lot like you learned how to walk. You try, awkwardly, one step at a time, falling, getting back up, and falling again. You keep doing it until your body learns the thousands of tiny dynamic adjustments necessary to keep you upright in every situation—so that you can take the next step without having to think at all.

Then you’re off to the races.

We also build AI tools for readers like you. Automate repeat writing with Spiral. Organize files automatically with Sparkle. Write something great with Lex.